Turn Your Data Into Smart Answers

Building AI-Powered Knowledge Bases

Part 1: Everything You Think You Know Is Probably Wrong

Most professionals hear "AI-powered knowledge base" and picture one of three things: a massive IT project that takes months, a glorified FAQ page with a chatbot slapped on top, or something only a company the size of Google can actually build. All three pictures are wrong. The real thing is closer to this: a marketing manager at a 12-person agency who spent one afternoon uploading her company's brand guidelines, past proposals, and client briefs into a tool called Notion AI, and now her team asks it questions instead of asking her. No developers. No budget approval. No six-month rollout. That story is repeating itself across industries right now, and the professionals who understand what's actually possible are pulling ahead fast.

Myth 1: You Need a Tech Team to Build One

The most common reason professionals don't build AI knowledge bases is the assumption that it requires developers, database architects, or at minimum someone who understands APIs. This belief made complete sense three years ago. Back then, connecting AI to your own documents genuinely required custom code, cloud infrastructure, and a budget that put it out of reach for most teams. That technical barrier is essentially gone. Tools like Notion AI, Microsoft Copilot for Microsoft 365, and ChatGPT's Projects feature now let you upload documents, connect folders, and build searchable, conversational knowledge systems entirely through point-and-click interfaces. The technical complexity has been absorbed into the product so you don't have to touch it.

Here's what a no-code knowledge base actually looks like in practice. A human resources manager at a mid-sized logistics company uploads her employee handbook, benefits documentation, onboarding checklists, and the last two years of policy update memos into a ChatGPT Project. She gives it a short set of instructions, essentially, "answer HR questions based on these documents, and tell employees when to contact HR directly for sensitive matters." From that point, her team can type questions like "What's our parental leave policy for part-time employees?" and get accurate, sourced answers in seconds. She built this in under two hours. No ticket to IT. No vendor contract. No coding.

Microsoft Copilot for Microsoft 365 takes this even further for teams already living in Word, Excel, SharePoint, and Teams. Because Copilot is embedded directly into those apps, it can reference your existing SharePoint document libraries, meeting transcripts, and email threads without you uploading anything manually. For a sales team, this means a new rep can ask Copilot "What objections did we handle in the Q3 enterprise deals?" and get a synthesized answer drawn from actual call notes and proposal documents stored in SharePoint. The knowledge base isn't something you build separately, it grows from the work your team is already doing.

Myth 2: An AI Knowledge Base Is Just a Fancy Search Bar

A traditional search bar finds documents that contain your keywords. You search "parental leave," you get five documents with those words in them, and then you read through each one to find the actual answer. An AI knowledge base does something categorically different: it reads and synthesizes across all those documents and gives you the answer directly, in plain language, with the ability to ask follow-up questions. This is not a marginal improvement. It's the difference between a filing cabinet and a knowledgeable colleague. The filing cabinet stores information. The colleague understands it and can explain it to you in context.

The synthesis capability is where real value appears. Imagine you're a consultant preparing a client proposal. You have 40 past proposals in a folder, research reports, competitor analyzis documents, and a set of pricing guidelines. A keyword search tells you which files mention "retail sector." An AI knowledge base lets you ask "What pricing approach have we used for retail clients with under $5M in revenue, and what were the outcomes?", and it synthesizes an answer from across all those documents simultaneously. That kind of cross-document reasoning used to require a senior analyzt spending half a day. Now it takes 30 seconds.

There's also the conversational dimension. You can ask a follow-up: "What about retail clients in the UK specifically?" The AI remembers the context of your first question and refines the answer. You can ask "Which of those proposals had the highest close rate?" and it keeps narrowing. This is closer to a research conversation than a search session. Tools like Claude Pro with uploaded documents, ChatGPT Projects, and Guru (a dedicated knowledge management platform with AI built in) all support this kind of multi-turn, contextual querying. Guru, specifically, is widely used by customer support and sales teams for exactly this reason, it surfaces relevant knowledge cards as employees work, rather than waiting for them to search.

Click to enlarge

Myth 3: Your Documents Have to Be Perfectly Organized First

This myth stops more people than almost any other. The thinking goes: "Our documents are a mess, different formats, inconsistent naming, some outdated, some duplicated. We need to clean everything up before we can use AI on it." So the knowledge base project gets added to the list of things that will happen after the big cleanup project, which never quite happens. Here's the corrected reality: modern AI tools are remarkably tolerant of messy, mixed-format document collections. ChatGPT Projects accepts PDFs, Word docs, spreadsheets, and plain text files simultaneously. Notion AI works across pages regardless of how they're structured. You don't need a taxonomy. You don't need consistent naming conventions. You don't need to purge outdated files before you start.

That said, quality still matters in one specific way: garbage in, garbage out. If you upload a policy document from 2019 that has since been replaced, the AI might surface outdated information without knowing it's outdated, because it has no way to know which version is current unless you tell it. The practical fix is simple: when you set up your knowledge base, include a short instruction like "If a document appears to conflict with another, flag both versions and note the dates." You can also simply delete or archive outdated files before uploading, which takes far less time than a full organizational overhaul. A focused 30-minute document triage is enough to get started.

Myth vs. Reality: The Full Picture

Click to enlarge

What Actually Works: The Mental Model You Need

Think of an AI knowledge base not as a piece of software you install, but as a new team member you onboard. When you hire someone, you give them context: here are our processes, here's our client history, here's how we handle complaints, here are the rules that matter. An AI knowledge base works the same way. You give it your documents (the context), you give it instructions about its role and boundaries (the job description), and then it's available 24/7 to answer questions, draft content, and surface relevant information for anyone on your team. The onboarding metaphor matters because it shifts your mindset from "building a system" to "briefing a resource", which is a task any professional already knows how to do.

The tools that make this most accessible right now fall into two categories. First, AI platforms with document upload and project features: ChatGPT Projects (available on ChatGPT Plus at $20/month) lets you upload up to 20 files per project and maintain persistent context across conversations. Claude Pro ($20/month) allows large document uploads and excels at careful, nuanced reading of complex materials like contracts, research reports, and long policy documents. Second, AI embedded in existing work platforms: Microsoft Copilot for Microsoft 365 (typically $30/user/month added to an existing Microsoft subscription) works directly inside Word, Excel, Teams, and SharePoint. Google Gemini with Google Workspace integration does the same for Drive, Docs, and Gmail. Notion AI ($10/month per member) is embedded directly in Notion workspaces.

The single most important thing to understand about what works: scope your first knowledge base narrowly. Don't try to capture everything your organization knows on day one. Pick one specific problem, onboarding new hires, answering client FAQs, supporting a sales team during proposal season, helping a customer service team handle product questions, and build for that. A focused knowledge base with 15 highly relevant documents outperforms a sprawling one with 200 loosely related files every time. You can always expand. The teams that succeed start with a clear question: "What does my team spend time looking up or explaining repeatedly?" That answer tells you exactly what to build first.

Click to enlarge

Frequently Asked Questions

• Can I use confidential company documents with these tools? This depends on your organization's data policies and the tool's privacy settings. ChatGPT Enterprise and Microsoft Copilot for Microsoft 365 offer enterprise-grade data privacy where your documents are not used to train AI models. ChatGPT Plus (the consumer version) has different terms. Always check with your IT or legal team before uploading sensitive client data, HR records, or proprietary financial information. When in doubt, use anonymized or non-sensitive versions of documents to test.
• How many documents can I upload? ChatGPT Projects currently supports up to 20 files per project, with a file size limit of 512MB per file. Claude Pro accepts very large documents, up to 100,000 words in a single upload, making it particularly useful for lengthy contracts or research reports. Notion AI has no separate upload limit since it reads your existing Notion workspace. Microsoft Copilot accesses your full SharePoint and OneDrive libraries with no additional upload required.
• Will the AI make things up from my documents? When AI is grounded in specific uploaded documents, what's technically called 'retrieval-augmented generation', it is significantly more accurate than when generating freely. It still isn't perfect. It can occasionally misread a table, miss a nuance, or blend information from two similar sections incorrectly. Always verify high-stakes answers against the source document. Building in a habit of spot-checking once a week keeps quality high.
• What file formats work? Most tools accept PDFs, Word documents (.docx), plain text files (.txt), and Markdown files. ChatGPT Projects also accepts Excel spreadsheets and PowerPoint files. Scanned PDFs (images of paper documents) are trickier, the AI can't read text that's embedded as an image rather than actual text. If you have scanned documents, convert them using a free tool like Adobe Acrobat's OCR feature or Google Drive's built-in document conversion before uploading.
• How do I keep the knowledge base up to date? Assign one person as the 'knowledge base owner', someone responsible for updating documents when policies change, new products launch, or processes shift. In practice, this means replacing outdated files with new versions every quarter, or whenever a significant change happens. Some teams add a 'last updated' note at the top of each document so the AI can reference it when answering questions about current versus past information.
• Can multiple people use the same knowledge base at once? Yes. ChatGPT Projects supports shared projects for teams. Microsoft Copilot is inherently multi-user since it connects to shared SharePoint libraries. Notion AI is accessible to everyone in your Notion workspace. Guru, a dedicated knowledge management platform used by companies like Slack and HubSpot, is specifically designed for team-wide knowledge access and includes features like content verification workflows and usage analytics to see which knowledge is actually being accessed.

Key Takeaways from Part 1

1. Building an AI knowledge base no longer requires developers, IT projects, or technical expertise, tools like ChatGPT Projects, Notion AI, and Microsoft Copilot make it a point-and-click process any professional can complete in hours.
2. An AI knowledge base is fundamentally different from search: it synthesizes across documents, answers in plain language, and supports conversational follow-up questions, capabilities that replace hours of manual research.
3. You don't need perfectly organized documents to start. A focused 30-minute triage of your most relevant files is enough. Scope your first knowledge base narrowly around one recurring problem.
4. The right mental model is onboarding a new team member, not installing software. You give the AI context (your documents) and a role (your instructions), and it becomes a 24/7 resource for your team.
5. Data privacy matters: know whether your tool's privacy settings are appropriate for the sensitivity of your documents before uploading anything confidential.

Three Myths That Are Holding Your Knowledge Base Back

Most professionals who've tried building an AI knowledge base hit a wall within the first two weeks. Not because the tools are too complicated, they're not, but because they're operating on assumptions that don't hold up. They dump documents in, expect magic, and get mediocre results. Then they blame the AI. The real problem is almost always one of three deeply held beliefs about how AI knowledge bases actually work. Each one is understandable. Each one is wrong. And correcting them changes everything about how you build and maintain your system.

Myth 1: More Documents Means Better Answers

The instinct makes sense. You want your AI assistant to be comprehensive, so you upload everything, the 2019 onboarding guide, six versions of the same policy document, a 200-slide deck from a conference, every email thread from the last product launch. The logic is: more information means fewer gaps. In reality, the opposite often happens. When an AI knowledge base is overloaded with redundant, outdated, or irrelevant content, it gets confused about which source to trust. It starts surfacing contradictory information. Users ask a simple question and get a hedged, muddled answer that cites three documents pointing in different directions.

Think of it like hiring a new employee and handing them a filing cabinet stuffed with every document your company has ever produced, including drafts, outdated procedures, and memos from five years ago. When someone asks them a question, they'll spend half their time sifting through irrelevant material. The same thing happens with AI. Tools like Notion AI, ChatGPT with file uploads, and Guru work best when the content they're drawing from is curated, current, and clearly organized. Quality of source material is the single biggest predictor of answer quality. This is not a technical limitation, it's a content strategy problem.

The professionals who build the most effective AI knowledge bases treat curation as an ongoing discipline, not a one-time upload. They designate someone, or a small team, to act as a knowledge curator. That person decides what goes in, what gets retired, and what needs to be rewritten before it's added. A marketing team at a mid-sized SaaS company reported cutting their knowledge base from 340 documents to 85 after an audit, and answer accuracy from their AI assistant jumped noticeably within the first week. Less clutter, sharper answers.

Myth 2: The AI Will Understand Your Documents Automatically

Upload a PDF, and the AI reads it, that part is true. But "reading" is not the same as "understanding context." AI tools process text. They don't automatically know that your document titled "Pricing_Final_v3" is actually outdated, that the "HR Policy" PDF refers to a location your company closed two years ago, or that a paragraph buried on page 14 of your employee handbook is the single most important piece of information for new hires. Without structure and context cues, the AI treats all text as equally relevant. That's where generic, unhelpful answers come from.

The fix is something knowledge management professionals call "metadata and framing." Before adding a document, you give it context, a clear title, a one-sentence description of what it's for, a tag indicating who it applies to, and a note about when it was last updated. In tools like Guru, Notion AI, and Confluence AI, this is built into the workflow. You're essentially writing a brief briefing note that sits alongside your document: "This is our current refund policy, updated March 2024, relevant to customer-facing teams." That framing dramatically improves how the AI retrieves and presents the information.

You also need to think about how your documents are written, not just what they contain. AI knowledge bases perform best with content that's structured in plain language, short paragraphs, clear headings, direct sentences. A 40-page legal contract is almost impossible for an AI to parse helpfully. But a one-page summary of the key terms and obligations, written in plain English with clear headers? That gets retrieved accurately almost every time. This is why the most effective knowledge base builders don't just upload existing documents, they often rewrite them first, or create AI-friendly summaries that sit alongside the originals.

Myth 3: Once Built, an AI Knowledge Base Runs Itself

This is the myth that causes the most expensive failures. Teams invest real time building a solid knowledge base, get good results in month one, and then treat it like a finished project. Nobody updates it when policies change. New product lines get launched without corresponding documentation. The AI keeps confidently answering questions using information that's now six months out of date, and users either stop trusting it or, worse, act on wrong information. A sales team that relies on an AI knowledge base with stale pricing data can quote clients incorrectly. An HR team with an outdated onboarding guide creates compliance risks.

An AI knowledge base is infrastructure, not a deliverable. The right mental model is a company intranet or a CRM, something that requires ongoing ownership, regular audits, and a clear process for updates. Best-practice teams assign a knowledge base owner (often a team lead or operations manager) who reviews content quarterly, flags documents approaching their "expiry date," and ensures new information gets added within a defined window, typically within one week of any policy or process change. Tools like Guru have built-in verification workflows that prompt document owners to confirm their content is still accurate on a scheduled basis. That kind of systematic maintenance is what separates knowledge bases that remain useful from ones that get quietly abandoned.

Click to enlarge

What Actually Works: The Practices Behind Effective AI Knowledge Bases

The teams getting consistent, reliable results from their AI knowledge bases share a few non-obvious habits. First, they write for retrieval, not for comprehensiveness. Every document added to the knowledge base is evaluated with one question: "If someone asked a question about this topic, would this document give a clear, direct answer?" If the answer is no, if the document is a 50-page contract, a sprawling strategy deck, or a meeting transcript, it either gets summarized before upload or stays out entirely. The knowledge base is not an archive. It's a curated answer engine.

Second, effective teams use test questions as a quality check. After adding or updating content, a designated team member asks the AI five to ten realiztic questions that users might ask about that content. If the answers are accurate, specific, and useful, the content passes. If the AI hedges, contradicts itself, or pulls from the wrong source, the content needs to be restructured. This takes fifteen minutes per document and prevents weeks of user frustration. It's the equivalent of proofreading, simple, often skipped, and critically important. Customer success teams at companies using Guru report running these "answer audits" monthly as standard practice.

Third, the best knowledge bases are built around actual user questions, not organizational chart logic. Most teams organize their knowledge base the way their company is structured. HR documents in one folder, Sales in another, Operations in a third. But users don't think in org-chart terms. They think in questions: "How do I submit an expense report?" "What's our policy on client gifts?" "Who approves contract extensions?" Building your knowledge base around the real questions your team asks, which you can collect through a simple form, a Slack channel, or just by asking, means the AI retrieves the right content because it was organized around retrieval from the start.

Click to enlarge

Frequently Asked Questions

• How many documents should I start with? Start with 15 to 25 documents maximum, specifically the ones your team references most often. You can identify these by asking colleagues what they search for most, or checking which files get opened most frequently in your shared drive. Build quality before you build volume.
• What file formats work best for AI knowledge bases? Plain text and well-structured Word or Google Docs files work best. PDFs work but perform less reliably, especially if they contain scanned images, complex tables, or multi-column layouts. If you must use a PDF, run it through a tool like Adobe Acrobat or Google Docs first to convert it to editable text, then clean up the formatting before uploading.
• Can I include confidential information in my knowledge base? This depends entirely on the tool and your organization's data policy. Tools like Microsoft Copilot for Microsoft 365 and Google Gemini for Workspace are designed to keep your data within your organization's environment. Consumer tools like ChatGPT free tier are not appropriate for confidential data. Always check your company's AI use policy before uploading sensitive documents.
• What if my team doesn't trust the AI's answers? Trust is built through transparency. Display the source document alongside every AI answer so users can verify it. Start with low-stakes use cases, answering internal FAQs, before using the knowledge base for client-facing or compliance-critical information. Trust grows as accuracy is demonstrated over time.
• How do I handle documents that change frequently? Create a 'living documents' tag or folder and schedule monthly reminders to review them. In tools like Guru, you can set automatic verification reminders that prompt the document owner to confirm or update content on a schedule. For very high-change content, like pricing or product specs, consider linking to a live source (a Google Sheet or internal dashboard) rather than uploading a static file.
• Do I need a special tool, or can I use ChatGPT? ChatGPT Plus (with file uploads) and Claude Pro work well for smaller, project-based knowledge bases where you're uploading documents for a specific conversation. For a persistent, team-wide knowledge base that multiple people access daily, purpose-built tools like Guru, Notion AI, or Microsoft Copilot for Microsoft 365 are better choices because they maintain the knowledge base between sessions and allow multi-user access with permissions.

Key Takeaways from Part 2

1. Curation beats volume every time, a smaller, cleaner knowledge base produces sharper, more trustworthy answers than a large, cluttered one.
2. AI doesn't automatically understand your documents, structure, plain language, and metadata are what make content retrievable and useful.
3. A knowledge base requires ongoing ownership, assign a specific person to manage updates, run quarterly audits, and keep content current.
4. Organize around user questions, not org-chart logic, the questions your team actually asks should determine how content is structured and labeled.
5. Test your knowledge base with real questions after every major update, fifteen minutes of answer-checking prevents weeks of user frustration and eroded trust.

What Most Professionals Get Wrong About AI Knowledge Bases

Most professionals believe that building an AI knowledge base requires a dedicated IT team, that the AI will understand their documents perfectly from day one, and that once it's set up, it runs itself. All three beliefs lead to real problems: wasted budgets, frustrated teams, and knowledge bases that quietly become useless over six months. The good news is that each misconception is fixable once you understand what's actually happening under the hood, without needing to touch a single line of code.

Myth 1: You Need IT to Build and Maintain This

The assumption that knowledge bases are IT infrastructure is understandable, for decades, they were. Enterprise wikis, intranets, and document management systems required server configuration, database schemas, and developer support. That model is dead for most professional use cases. Tools like Notion AI, Microsoft Copilot integrated with SharePoint, and platforms like Guru or Tettra are built specifically so that a marketing manager or HR director can set up a fully functional AI-powered knowledge base in an afternoon, using drag-and-drop interfaces and plain-language configuration.

The real role IT plays today is narrower: data security review, single sign-on integration, and enterprise licensing. Those are important conversations to have, but they don't block you from starting. You can prototype your knowledge base with free tools, demonstrate value to your team, and then loop in IT when you're ready to scale or connect it to sensitive systems. Starting is a business decision, not a technical one.

Click to enlarge

Myth 2: The AI Understands Your Documents Automatically

Upload a PDF and the AI will just 'get it', this is the most dangerous myth. AI knowledge base tools read text. They do not interpret poorly formatted tables, decode scanned images, understand your internal acronyms, or know that 'Q3 target' means something different in your sales team than your finance team. A knowledge base is only as good as the quality of what you put into it. Garbage in, confident-but-wrong answers out. The AI will answer questions with apparent authority even when the source material is ambiguous or incomplete.

The fix is document hygiene before upload. That means converting scanned PDFs to text-searchable versions (Adobe Acrobat's free online tool does this in seconds), breaking long documents into clearly titled sections, and adding brief summaries at the top of each document. Think of it like briefing a very smart new hire: they need structure and context to give useful answers. A 40-page policy document with no headings is harder for the AI to work with than ten short, well-labeled pages covering the same content.

Acronyms and internal jargon are a particular trap. If your company calls its performance review process 'the Pulse' and your documents never explain what that means, the AI will either ignore the term or guess incorrectly. Maintain a simple glossary document in your knowledge base, a one-page list of internal terms and their meanings. Upload it alongside everything else. This single step dramatically improves answer accuracy for any team with its own vocabulary.

Myth 3: Set It and Forget It

An AI knowledge base is not a database you configure once and walk away from. It reflects whatever documents live inside it, and documents go stale. A benefits guide from 2022 will confidently tell employees about a health plan that no longer exists. A sales playbook from last year will reference a pricing tier you discontinued in March. The AI doesn't know what it doesn't know. It will answer from outdated material with exactly the same confidence it answers from current material. This is not a flaw in the AI, it's a content management problem that every knowledge base has always had. AI just makes the wrong answer sound more authoritative.

The solution is a simple quarterly review ritual, not a technical process. Assign document ownership, each document in your knowledge base has a named person responsible for keeping it current. Set a calendar reminder every 90 days to review and re-upload updated versions. Some platforms like Guru have built-in 'verification' workflows that automatically flag documents older than a set period and ask the owner to confirm they're still accurate. That feature alone is worth the subscription cost for teams with more than 20 documents.

What Actually Works: The Practical Model

The teams that get real value from AI knowledge bases treat them less like software and more like a living reference library with a very fast librarian. They start small, 10 to 20 well-formatted documents covering the questions their team asks most often. They test the knowledge base themselves before sharing it, using the kinds of questions real colleagues ask. They fix gaps in the source documents rather than trying to fix the AI. And they expand gradually, adding new document categories only after the existing ones are working reliably.

The tools that work best for non-technical teams right now are Notion AI for internal wikis and project knowledge, Microsoft Copilot for organizations already using Microsoft 365, Guru for customer-facing and sales teams, and ChatGPT Team or Claude for ad hoc document Q&A. None of these require configuration beyond uploading files and setting who can access what. The choice between them usually comes down to where your documents already live, use the tool that connects most naturally to your existing workflow.

The biggest predictor of success isn't the tool you choose, it's whether someone owns the knowledge base. Teams that assign a specific person (even part-time) to manage document quality, respond to user feedback, and run quarterly reviews consistently outperform teams that treat the knowledge base as shared infrastructure with no clear owner. This is a people and process decision, not a technology one. The AI does the retrieval. Humans have to do the curation.

Frequently Asked Questions

• Q: Can I use confidential HR or legal documents in an AI knowledge base? A: Yes, but choose tools with enterprise-grade security and check your organization's data policy first. Microsoft Copilot with SharePoint and Guru both offer data isolation. Avoid uploading sensitive documents to free-tier consumer tools like the free version of ChatGPT.
• Q: How many documents can a knowledge base handle before it gets slower or less accurate? A: Most no-code platforms handle hundreds of documents without performance issues. Accuracy doesn't degrade with volume, it degrades with poor document quality. A well-structured library of 200 documents outperforms a messy library of 20.
• Q: What file formats work best? A: Plain text, Word documents (.docx), and text-searchable PDFs work best. Scanned image PDFs, heavily formatted spreadsheets, and PowerPoint files with mostly graphics tend to produce poor results. Convert or reformat before uploading.
• Q: Do I need to retrain the AI when I update a document? A: No. With no-code tools, you simply delete the old version and upload the new one. The AI reads the current documents in the knowledge base, there's no separate training step required.
• Q: What if the AI gives a wrong answer and someone acts on it? A: This is a real risk, which is why testing before launch matters. Add a visible disclaimer to your knowledge base interface: 'Always verify critical decisions with the original policy document or your manager.' Use the knowledge base to speed up finding information, not to replace human judgment on high-stakes decisions.
• Q: How do I get my team to actually use it instead of just emailing me questions? A: Surface the knowledge base where questions currently happen. If your team uses Slack, add a channel pinned message linking to it. If they use Teams, add it as a tab. Habit change follows friction reduction, make the knowledge base the path of least resistance, and adoption follows.

Key Takeaways

• No-code tools like Notion AI, Guru, and Microsoft Copilot let any professional build an AI knowledge base without IT support or technical skills.
• AI doesn't automatically understand your documents, answer quality depends on how well-structured, clearly written, and up-to-date your source material is.
• Document hygiene is the most important maintenance task: clear headings, plain language, a glossary of internal terms, and text-searchable file formats.
• AI knowledge bases go stale if no one owns them. Assign a document owner and schedule quarterly reviews to keep content accurate.
• Test your knowledge base with realiztic questions before sharing it. Fix gaps in the source documents, not in the AI settings.
• Start with your top 10 FAQs, prove the value, and expand gradually. A small, accurate knowledge base beats a large, unreliable one every time.